Buckets:
| #!/usr/bin/env python | |
| """Figure: Observability Paradox + tail stability (Claims 1 & 2), REPORTED. | |
| Bar chart of the paper's reported Table 4 mean and p95 relative-to-best gaps | |
| across observability/reasoning configs. Clearly labelled REPORTED (the LLM | |
| slice was not run here). Highlights L2 CoT (best mean + tightest tail) vs | |
| L3 CoT (worse despite full structural priors) = the Observability Paradox. | |
| """ | |
| from __future__ import annotations | |
| import json | |
| from pathlib import Path | |
| import matplotlib | |
| matplotlib.use("Agg") | |
| import matplotlib.pyplot as plt # noqa: E402 | |
| import numpy as np # noqa: E402 | |
| ROOT = Path("/home/ubuntu/samuel/dynasched-repro") | |
| SRC = ROOT / "outputs" / "observability_reported.json" | |
| OUT = ROOT / "figs" / "observability_l2_vs_l3.png" | |
| def main(): | |
| d = json.loads(SRC.read_text())["table4_reported"] | |
| configs = ["L1 Direct", "L1+Tool", "L2 CoT", "L3 CoT"] | |
| means = [d[c]["mean"] for c in configs] | |
| p95 = [d[c]["p95"] for c in configs] | |
| x = np.arange(len(configs)) | |
| w = 0.38 | |
| colors_mean = ["#94a3b8", "#94a3b8", "#0f766e", "#b91c1c"] | |
| colors_p95 = ["#cbd5e1", "#cbd5e1", "#5eead4", "#fca5a5"] | |
| fig, ax = plt.subplots(figsize=(9, 5.5)) | |
| b1 = ax.bar(x - w / 2, means, w, color=colors_mean, label="mean gap %") | |
| b2 = ax.bar(x + w / 2, p95, w, color=colors_p95, label="p95 gap %") | |
| ax.set_xticks(x) | |
| ax.set_xticklabels(configs) | |
| ax.set_ylabel("Relative-to-best makespan gap (%)") | |
| ax.set_title( | |
| "Observability Paradox (Table 4, REPORTED — LLM slice not run)\n" | |
| "L2 CoT beats L3 CoT on both mean (0.65<1.66) and p95 tail (2.01<2.92)" | |
| ) | |
| for bars in (b1, b2): | |
| for bar in bars: | |
| ax.text( | |
| bar.get_x() + bar.get_width() / 2, | |
| bar.get_height() + 0.05, | |
| f"{bar.get_height():.2f}", | |
| ha="center", | |
| va="bottom", | |
| fontsize=8, | |
| ) | |
| ax.legend() | |
| ax.text( | |
| 0.99, | |
| 0.97, | |
| "REPORTED — not reproduced (no LLM rollouts)", | |
| transform=ax.transAxes, | |
| ha="right", | |
| va="top", | |
| fontsize=9, | |
| color="#b91c1c", | |
| style="italic", | |
| ) | |
| plt.tight_layout() | |
| OUT.parent.mkdir(parents=True, exist_ok=True) | |
| fig.savefig(OUT, dpi=130) | |
| print(f"Wrote {OUT}") | |
| if __name__ == "__main__": | |
| main() | |
Xet Storage Details
- Size:
- 2.36 kB
- Xet hash:
- f32fd6e96faa474c22a2540d19405974f7a94d18f12af73ed5c0fc6a75ed3742
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.